LM Studio 0.4.0 Released
LM Studio is an AI tool that is in regular rotation in my personal environments. I like LM Studio because it lets me experiment with models on my own hardware. Although this is sometimes a constraint for me due to what hardware I have available. After eighteen beta builds, the LM Studio team has released version 0.4.0. This release is a significant advancement that transitions the product from a primarily desktop-focused application into a flexible local inference platform suitable for developers, teams, and lightweight production environments.This release combines architectural, usability, and reliability improvements into a cohesive upgrade.
A central feature of the release is a new headless architecture with the introduction of a background daemon named llmster. This component enables LM Studio to operate in headless mode without a GUI and supports server-style or automated deployments. This is complemented by support for parallel inference requests using continuous batching. This capability allows the system to process multiple prompts simultaneously rather than sequentially, which improves throughput. A new stateful REST API expands integration options, making it easier for external tools and applications to interact with local models, including workflows based on a MCP. Permission tokens are available to secure access to the API.
The user experience has been refreshed with a redesigned interface that focuses on productivity and discoverability. Significant enhancements include a split-view chat for side-by-side conversations, a unified developer mode that gathers advanced capabilities in one location, and an improved model search experience with persistent filters and better navigation. These changes reduce friction for both casual users and power users who manage multiple models and workflows.
For developers, LM Studio 0.4.0 introduces improvements to the CLI, including a more capable terminal-based chat experience and refined tools for runtime inspection and versioning. Combined with the new daemon and API capabilities, these enhancements provide a foundation for scripting, automation, and integration into local development pipelines.
Finally, the release includes stability, performance, and usability fixes developed throughout the later builds. These updates include improvements to vision model handling, MCP loading behavior, settings persistence, token counting, and API reliability. These fixes address day-to-day reliability and resolve user interface inconsistencies. I am currently in the process of downloading and installing version 0.4.0 to begin another round of GenAI experiments this weekend. I should probably download the Python SDK as well.
